An OT Approach to Phonology-Syntax Mappings in Chinese Tone Sandhi
Wei-wen Roger Liao
USC Linguistics
1.
Introduction
This paper deals with an old puzzle in a new approach. The formal properties of Tone Sandhi (TS)
in Chinese are widely studied in the literature of generative phonology. It has been argued that
TS in several dialects of Chinese patterns with syntactic structures, while some also argue that
this syntax-phonology mapping is not that transparent (see Chen 2000: Ch.7 and Ch.10 for
extensive surveys). In this paper, syntax-phonology mappings of TS in several Chinese dialects
are studied in a constraint-based OT approach (Prince and Smolensky 1993). I argue for a direct
phonology-syntax mapping relation between dominance of syntactic objects and phonological
constraints.
Among studies of universal syntax-phonology mappings, a widely received view is that
prosodic structures in phonology are shaped by maximal projections in syntax (i.e. XP/Xmax).
This is generally referred to as the XP-to-P Mapping Condition (see Selkirk 1986, 1995;
Truckenbrodt 1999, and the references listed there). However, I show this XP-to-P Mapping
Condition is either too strong or too weak in predicting the TS patterns in Mandarin Chinese. On
the other hand, I argue that syntactic dominance relations are the sole factor linking
phonology-syntax mappings. Technically, I propose a set of phonology-syntax mapping
constraints, which are able to make reference to syntactic dominance, and the dominance
relations can be directly translated into the rankings of these mapping constraints. The proposed
direct phonology-syntax mapping has the following mechanisms: in the structure [XP ab [YP cd
[ZP ef]]] (abcdef are phonological representations), where the string contained by XP is (abcdef),
the one by YP is (cdef), and the one by ZP (ef), we predict that the ranking among CON(XP),
CON(YP), and CON (ZP) is CON(XP) >> CON(YP) >> CON (ZP) (CON(Σ) for some constraint
targeting a syntactic domain Σ). On the other hand, a ranking which does not follow the
dominance relation is immediately ruled out in this system (i.e. * CON (ZP) >> CON(XP) >> CON
(YP)). Optimality theoretically, the constraint violations caused by the string (abcdef), if any, are
considered more unacceptable than those by the string (cdef), which, in turn, are more
unacceptable than those by the string (ef). In addition, since dominance relations are accessible to
OT constraints, we predict that there are constraints which may refer to the most/least dominant
syntactic objects, represented as CON(ΣMAX) or CON(ΣMIN). CON(ΣMAX) targets the strings
contained by the most dominant syntactic objects, while CON(ΣMIN) the smallest strings by the
least dominant syntactic objects.
Applying the formal mechanisms to the analysis of Mandarin TS, I will show that TS in
Mandarin can be analyzed as competitions between a set of faithfulness constraints (ANCHORING)
which targets right edges of different syntactic domains. A tendency is observed in Mandarin TS:
the more dominant a syntactic domain is, the more likely its right edge to remain faithful. With
the help of the direct mapping mechanisms, the generalization can now be captured by
phonology-syntax mapping constraints, such as in the construction [XP X [YP Y]], ANCHORING
(XP, Right) ranks over ANCHORING (YP, Right) (the constraints will be defined later). Hence the
shift of tone in the right edge of XP will be considered worse than that in the right edge of YP.
Furthermore, from an asymmetry between leftward and rightward compounds, I propose that an
anti-faithfulness constraint, GREED(SMIN), is needed, which also relies on syntactic imports. In
Mandarin, GREED targets the least dominant strings SMIN and forces the exhaustive TS in those
domains. Combining the results, the final ranking in Mandarin will be a set of hierarchically
mapping ANCHORING constraints dominating the GREED constraint, and they in turn dominate a
faithfulness IDENT constraint.
2
As evidence for the proposed direct mapping condition, I look into two other main
dialects of Chinese, Taiwanese (or Xiamen) and Shanghai Chinese, as well as a native American
language, Tohono O’odham (or Papago). I will show the same direct phonology-syntax
mappings can be used to describe the tonal patterns in these languages. For instance, in
Taiwanese, GREED targets the largest string(s), i.e., GREED(SMAX); therefore, TS applies
exhaustively within the largest string(s). The same operations on maximal strings can be found in
Tohono O’odham. On the other hand, Shanghai Chinese and Mandarin share similar constraint
ranking. The structural hierarchy is also mapped to the rankings of ANCHORING constraints. The
differences being that in Shanghai Chinese, ANCHORING targets left edges, the mirror image
between Shanghai and Mandarin is therefore accounted for.
This paper is organized as follows: General tonal patterns and TS rules in Mandarin are
discussed in section 2.1, in which three asymmetries of TS are observed across several syntactic
domains. After a short review of earlier proposals by Cheng (1987) and by Truckenbrodt (1999)
in section 2.2, I analyze TS in Mandarin in a constraint-based theory in section 3. The proposed
ranking in section 3, however, finds exceptions in cases involving subjects and higher adverbs. In
section 4, I show that the same problems from subject and higher adverbs can also be found in
Taiwanese TS, and these problems in Mandarin and in Taiwanese TS turn out to provide
evidence for the claim that major prosodic boundaries are regulated by multiple spell-outs
(Kratzer & Selkirk 2007; Simpson & Wu 2002). Once multiple spell-outs are introduced in the
system, subject and higher adverbs no longer pose threats to the current analysis. In section 5, I
examine tone spreading in Shanghai Chinese and tone grouping in Tohono O’odham. I show that
the same direct phonology-syntax mapping mechanisms can deal with various tonal patterns in
these languages.
3
2.
Tone Sandhi in Mandarin
2.1
Three Asymmetries from Syntax Mapping
In Mandarin, TS applies when two (underlying) Low tones occur adjacent to each other.1 An
underlying Low-Low (LL) sequence becomes Rising-Low (RL) on the surface representation. As
observed in many studies, TS is sensitive to certain syntactic domains (Chen 2000; Cheng 1987;
Shih 1986, among others). The following examples illustrate the syntactic effects in word levels
and sentence levels:
(1) Word-internal (= for compound boundaries; - for morpheme boundaries; # for TS
boundaries)
a. zong-tong
‘president’
L L
UT (underlying tone)
R L
ST (surface tone)
b.
zong-tong=fu
L L
L
R R
L
* L# R
L
‘the president’s house’ (rightward compounds; 2+1)
UT
ST
c.
lao=zong-tong
L L L
L# R L
* R R L
‘the senior president’
UT
ST
(leftward compounds; 1+2)
(1a) shows a basic TS pattern, where two adjacent Low tones (LL) become Rising-Low (RL).
Note that the right edge (L) in every example is well-preserved. (1b) is an instance of rightward
compounds (the disyllabic zong-tong and the monosyllabic fu). The priority to preserve the right
edge arguably drives TS of the middle word tong. However, the leftmost low tone still undergoes
TS although the undesired LL sequence would not be created if TS did not apply (i.e. LRL). This
1
The low tone is often transcribed in textbooks as a low-falling-rising tone in standard Mandarin. However, in daily
speech, we can transcribe it as low. This point, however, does not affect the validity of the proposed analysis in any
way.
4
over-application of TS indicates that TS is somehow obligatory to the first two low tones. On the
other hand, in the leftward compounds (a monosyllabic word and a disyllabic word), as in (1c),
there seems to be a TS boundary which blocks the over-application of TS to the first two
syllables. Obviously, there is an asymmetry between leftward and rightward compounds. I will
refer to this as a ‘directional asymmetry in compounds’.
One way of looking at the directional asymmetry is to claim that (word-internal) syntax is
involved in TS applications. Suppose compounds in (1b) and (1c) have corresponding syntactic
structures in the following:
(2)
a.=(1b)
N
N
b.=(1c)
N
N
A
zong-tong
fu
lao
LL
( RL)
( RR
(L)
L
L)
L
L
(L
N
zong-tong
LL
(RL)
RL)
UT
ST
Shih (1986) and Cheng (1987) claim that the ordering of applying TS rules (LL  RL) is subject
to cyclic groupings in syntax, illustrated in (2). We will come back to Cheng’s proposal in 2.2.
Nevertheless, this example illustrates that the applications of TS are very likely to be shaped by
corresponding syntactic structures. A descriptive generalization drawn from this example is a
tendency that TS applies more exhaustively to the lower strings, while the higher right edges are
more likely to be preserved:
5
(3) A Tendency in the Syntax-Phonology Mappings of Mandarin TS
The higher a syntactic domain is, the more likely the underlying tone in its right edge is
preserved; the lower a syntactic domain is, the more likely the low-tone strings (in that
domain) undergo exhaustive TS.
The same tendency is observed when TS applies in sentences, such as in the following examples.
In (4a) TS applies exhaustively to one of the least dominant domains [wu-bai] ‘five-hundred’,
while the highest and rightmost edge, which is occupied by li ‘kilometer’ remains its underlying
low tone:
(4)
a. [Lao-ma [pao
[[wu-bai]
li]]]
name.
run
five-hundred
kilometer
L L L
L L
L
R R L
R R
L
R L# L
R R
L
* R R L # L R
L
‘Laoma runs for five-hundred kilometers.’
b. [Xiao-hu [da
[lao-shu]]]
name
hit
mouse
L L L
L L
R R L
R L
R L# L
R L
?R L R
R L
‘Xiaohu hit the mouse.’
UT
ST1 (unmarked)
ST2 (unmarked)
ST3 (unacceptable)
UT
ST1 (unmarked)
ST2 (unmarked)
ST3 (marked)
In addition to the same tendency, another asymmetry can be observed. In both (4a) and (4b), a
TS boundary can be optionally inserted between the subject and the main verb. Therefore, ST1
and ST2 in both examples are equally unmarked to most native speakers although there is an
undesirable LL sequence created by the subject and the verb (a subject does not need to be
pronounced with an emphatic stress, and ST1 is preferred in fast speech). On the other hand, no
such boundaries can ever appear between a verb and its object, as shown in ST3 in (4b). I will
6
call this second asymmetry as a ‘subject/object asymmetry.’
A third asymmetry in the phonology-syntax mappings of Mandarin TS involves adverbs
of different heights:
(5) a.
b.
Lao-ma [hao-hao jiang]
name.
well
speak
L L L L
L
R L R R
L
* R L R
L# L
‘Laoma speaks in a good manner.’
[lower adverb]
UT
ST (unmarked)
ST (marginal)
Lao-ma zheng-zheng [pao [wu bai
li]]
[higher adverb]
name.
completely
run
five hundred kilometer
L L L
L
L
L
L
L
UT
R L R
L
# L
R
R
L
ST1 (unmarked)
R L R
R
L
R
R
L
ST2 (unmarked)
R R R
R
L
R
R
L
ST3 (less unmarked)
‘Laoma ran five hundred kilometers completely.’
Similar to the subject/object asymmetry, an LL sequence is well tolerated between a higher
adverb and a verb, which indicates that there is a TS boundary in between, as shown in (5b). On
the contrary, a TS boundary cannot appear in between a lower adverb and a verb, as shown in
(5a). The asymmetric tonal patterns can be attributed to different syntactic positions of the
adverbs. Let us adopt the idea that lower adverbs are VP-based and higher adverbs are
IP/CP-based (Jackendoff 1972; Cinque 1999); therefore, lower adverbs, being closer to the verb,
are always grouped with the verb when TS occurs. On the other hand, higher adverbs (behaving
more like subjects), can escape from a TS group more easily. This asymmetry will be referred to
as the ‘higher/lower adverb asymmetry.’
7
In summary, TS in Mandarin is sensitive to syntax. In terms of syntactic hierarchy, there is a
tendency that right edges of the more dominant strings resist changes, while the least dominant
strings will undergo exhaustive TS. In addition, we see three asymmetries in Mandarin TS,
which are also related to syntactic terms: (i) the directional asymmetry in compounds, (ii) the
subject/object asymmetry, and (iii) the higher/lower adverb asymmetry. A well-rounded analysis
of Mandarin TS, then, should not only be able to capture the phonology-syntax mappings, but to
account for all of these asymmetries.
2.2
Previous Analyses
2.2.1 The Prosody Hierarchy: Cheng (1987)
Following Hayes (1984/1989), Cheng (1987) proposes that TS is subject to a prosodic hierarchy.
Cheng distinguishes three levels of prosodic hierarchies in Mandarin TS, word-level,
Phonological-Phrasing (PhP), and Intonational-Phrasing (InP). The system is rule-based, and the
TS rule states that each of the non-final low tones becomes a rising tone when a series of low
tones occur in the same phrasing. An example can be drawn in (6):
(6) ‘Xiaomei invited Xiaohu to hit the mouse.’
Xiao-mei
Name
L L
( R L)
( R L)
(( R R
R R
zhao
ask
L
L
(L
L
L
xiao-hu
Name
L L
( R L)
R L)
R R)
R R
da
hit
L
L
(L
L
L
lao-shu
mouse
L L
( R L)
R L)
R L)
R L
8
UT
word-level phrasing
PhP
InP
ST
Some of the generalizations are indeed well captured in Cheng’s model. First, TS boundaries are
generally mapped from syntax. Therefore, the low tones in the dominant right edges are more
likely to be preserved.
However, the approach is not without problems. First, Cheng’s analysis cannot account
for the optional tone preservations in the right edges of subjects and higher adverbs. As discussed
earlier in the subject/object asymmetry and the higher/lower adverb asymmetry, a TS boundary
may appear in these positions (after subjects and higher adverbs) optionally. Cheng’s model is
considered to be too strong on this ground because her model will predict that a subject or a
higher adverb will not introduce a TS boundary (ultimately, the subject/higher adverb and VP
will be grouped in the same intonational phrasing). Second, Cheng’s system uses an indirect
mapping between syntax and phonology (only phonological-phrasing is mapped from syntactic
XPs). It is not clear whether syntax also plays a role in word-levels and intonation-levels. This
model thus leads to a conceptual disadvantage. Namely, too many levels of computations are
involved across different domains and different linguistic modules. The question we can ask is
whether there is a way to deal with TS in a conceptually simpler once-and-for-all fashion, as
generally assumed in optimality theory. Following this line, if syntax really influences TS in an
apparent way, we should also expect, in a conceptually superior system, a uniform source of
phonology-syntax mappings, which I will argue, is the dominance relations in syntax.
2.2.2 WRAP-XP VS. ALIGN-XP: Truckenbrodt (1999)
Although Truckenbrodt (1999) does not discuss TS in Mandarin, he tries to deal with the general
syntax-phonology mappings (in other tonal languages) in an OT approach. Specifically, he
proposes a WRAP-XP constraint and an ALIGN-XP constraint to capture the XP-to-P mapping
condition, proposed in Selkirk (1986):
9
(7)
XP-to-P Mapping Condition (from Truckenbrodt 1999: 221)
Mapping constraints relate XPs to phonological phrases, but do not relate XPs to other
prosodic entities.
The constraint WRAP-XP requires that XPs are contained in a single p-phrasing (hence extending
p-phrasing if an XP dominates another XP), while the Align-XP requires that every XP is aligned
with a p-phrasing. By definition the two constraints have a competitive nature. The interactions
between WRAP-XP and ALIGN-XP can be shown in the following illustrations:
(8)
The competitions between WRAP-XP and ALIGN (XP, RIGHT) (# for p-boundary)
a. WRAP-XP >> ALIGN (XP, RT):
(i) [AP A [BP B]]

(ii) [AP [ZP Z ] [AP A [BP B]]] 
(AB)#
(ZAB)#
b. ALIGN (XP, RT) >> WRAP-XP:
(i) [AP A [BP B]]

(ii) [AP [ZP Z] [AP A [BP B]]] 
(A(B))#
(Z)#(A(B))#
(AP and BP are contained in a P)
(AP, BP, and ZP are contained in a P)
(AP and BP are each aligned with a P)
(AP, BP, and ZP are each aligned with a P)
Since Mandarin is generally head-initial, the rankings in (8a) and (8b) should predict the same
TS patterns when there are no adjuncts. When there is an adjunct (ZP), the ranking in (8a)
predicts that everything is phrased together in one p-phrasing, while the ranking in (8b) predicts
that the adjunct itself will introduces a p-boundary.2 Let us apply this analysis to TS in Mandarin.
First, in head-complement structures, we correctly predict that a head and its complement do not
have a TS boundary in between no matter which ranking obtains. However, this is probably the
only case where this theory makes correct predictions. Suppose A=T/I (so AP licenses the
subject), we incorrectly predict that the subject and VP would always have no TS boundary since
a subject is not an adjunct. Suppose ZP is a higher adverb, the theory fails to predict the optional
TS boundary, either (no matter which ranking is adopted). Truckenbrodt (1999) circumvents
2
Truckenbrodt (1999) assumes that in adjunction structures, the higher segment is not an XP itself.
10
these potential problems by postulating a Lexical Category Condition (see Truckenbrodt 1999:
226), in which the constraints apply to lexical categories only. Therefore, since higher adverbs
and subjects are licensed by functional categories, he may argue that they are immune to
WRAP-XP and ALIGN-XP. But even so, a more precise explanation is called for why functional
categories (and their branches) can easily escape the XP-to-P condition, and it is not very clear
whether Truckenbrodt has offered a complete analysis to these functional categories.3 A more
serious problem for Truckenbrodt’s analysis comes from the directional asymmetry of
compounds in Mandarin TS. In compounds, arguably there are no XPs involved. Therefore, we
should expect that in compounds, a TS boundary will never be introduced. The theory thus
predicts that in word levels, directions in compounds do not make any different predictions on
tonal patterns, contrary to the facts observed in Mandarin TS.
3.
An OT Analysis: Direct Phonology-Syntax Mapping
In this section, first I define the primitives and constraints in the proposed system. In 3.2, various
patterns in Mandarin TS are discussed in details, along with applications of the direct
syntax-phonology mappings.
3.1
Basic Architecture
As a general condition, for a syntactic object Σ = XP or X, where XP is a maximal projection of
its head X, Σ contains a phonological string S. For convenience, I adopt the following notations:
[…] for an XP domain, and <…> for an X domain. The structure in (9) illustrates a simple case,
where a monosyllabic verb takes a disyllabic object NP:4
3
In Truckenbrodt’s analysis of Tohono O’odham, sometimes IP introduces a p-boundary (his (18)), and sometimes
IP does not (his (19) and (20)). It does not seem to be the case that the p-boundary is optional here, and
Truckenbrodt does make a clear explanation on this point.
4
The monosyllabic verb itself takes a string consisting of one syllable, but this monosyllabic string is not visible to
tone sandhi since it is not a minimal rhythmic unit (Chen 2000), which requires a string to include at least two
syllables (a syllable is always a tone bearing unit in Mandarin). Therefore, in (9), the monosyllabic string is ignored
in the discussion.
11
(9)
a.
VP: [qing [<zong-tong>1]2]3
‘ask the president’
b.
VP, S3
V
Σ1 = N = S1 = <LL>
Σ2 = NP = S2 = [LL]
Σ3 = VP = S3 = [LLL]
NP, S2
N, S1
qing
[L
zong-tong
[<L L>]]
When two phonological strings are put together at syntax, they automatically create a larger unit
in which both strings are contained. This larger unit either immediately dominates the two strings,
or in the case of adjunctions, the two strings will be equally dominant. The structural dominance
is defined as follows (adapted from Chomsky 1986):
(10)
X dominates Y iff (a) X and Y are categories and X is a higher node than Y, or (b) X is a
segment and every segment of X is a higher node than Y.
By definition, dominance is an asymmetric, partial, and transitive relation (McCawley 1968).
This means when two phonological strings, each held by a syntactic object, are combined at
syntax, they will also inherit the properties of being asymmetric, partial, and transitive. Take (9)
as an example, the dominance relations between the phonological strings mapped from syntax
are in (11):
(11)
Dominance relations (>) between syntactic objects and phonological strings
VP > NP > N ↔ Σ3 > Σ2 > Σ1 ↔ S3 > S2 > S1
This translation between phonological strings and syntactic objects will be the single source of
phonology-syntax mapping in the proposed system, the applications of which will soon be
12
discussed.
Proceed to the constraints. First, we observe that the LL sequence is generally banned in
Mandarin, which can be viewed as an OCP effect that banned the consecutive low tones in
Mandarin. The markedness constraint is defined as follows:
(C1) OCP(L)
A series of two consecutive low tones is prohibited (i.e. *LL).
Recall the tendency from section 2. The first part is that right edges in more dominant domains
are more likely to be preserved. The second part is that lowest domains undergo exhaustive TS.
For the earlier, we may formulate a faithfulness constraint which requires preservations of the
right edges in certain syntactic domains:
(C2) ANCHORING -IO(Si, RIGHT)
The right edge of Si (mapped from Σi) in an output corresponds to the right edge of Si in
the input.
The dominance relations between different syntactic domains are directly translated into the
dominance of the anchoring constraints targeting syntactic domains. This direct mapping
between phonology and syntax is formulated as follows:
(12) The Direct Phonology-Syntax Hypothesis
The ranking between CON(Σi) and CON (Σj) is mapped from the syntactic dominance
between Σi and Σj.
As for the latter part of the tendency (lowest strings undergo exhaustive TS), it can be
formulated by an anti-faithfulness constraint (Alderete 2001), which maximizes TS in a certain
13
syntactic domain. By nature of this constraint, I will call this constraint GREED. In Mandarin TS,
GREED focuses on the low tones and targets the least dominant syntactic domains. The general
form of this constraint is formulated in (C3):
(C3) GREED (T, ΣMIN/MAX) = ~IDENT-IO (T, ΣMIN/MAX)
Maximize the differences between the input and output strings with respect to the
specified tone(s) inside the least/most dominant syntactic domains (T for specified
tone(s)).
(e.g. in Mandarin: GREED (L, ΣMIN))
To prevent over-applications of TS outside the minimal domains, we need another faithfulness
constraint, which does not restrict its focus on the right edge, but on the whole phonological
string. With this faithfulness constraint, we wish to avoid any over-applications except those
driven by GREED:
(C4) IDENT-IO
Maximize the correspondences between the input and output strings.
Prima facie, it seems that OCP(L) and GREED (L, ΣMIN) apply redundantly. However, closer
examinations indicate that the two constraints are independently motivated, and they apply to
different modules of grammar (although they may overlap in some cases).5 First of all, OCP(L)
punishes two adjacent low tones in a string. It is motivated by a language-specific markedness
pattern (*LL in Mandarin). On the other hand, GREED (L, ΣMIN) is not motivated by any
markedness patterns; it simply requires low tones to undergo TS when it occurs inside certain
syntactic domains. Obviously, this requirement in shifting the tonal patterns is not solely
phonologically driven.6 Given syntactic information, the two constraints may actually prefer
5
The overlapping in the domains of application, however, does not pose a problem for OT.
It remains unknown what drives TS, if not for phonological reasons. However, as we will see later in Taiwanese
TS, markedness tonal patterns are not enough to drive TS (since in this languages, TS occurs exhaustively to every
6
14
different surface tonal patterns, as seen in the following charts:
(13) The Division of Labor between the Antifaithfulness GREED and the Markedness OCP
GREED(ΣMIN)
OCP
a. Input: <LL>
RR
RL, LR, RR
b. Input: <LLL>
RRR
RLR, LRL, RRR, RRL, LRR
c. Input: <<LL>L>
RRL
RLR, LRL, RRR, RRL, LRR
d. Input: <L<LL>>
LRR
RLR, LRL, RRR. RRL, LRR
e. Input: <L<L<LL>>>
LLRR
LRLR, LRRL, LRRR, RRRL,
RRRR, RRLR, RLRR, RLRL
From (13), we see OCP itself does not force the exhaustive TS, but this is the job of GREED, as in
(13a) and (13b). Also, only GREED is concerned with syntactic information, being a
phonology-syntax mapping constraint. On the contrary, being a pure phonological constraint,
syntactic information is totally invisible by OCP. (13e) shows a case where LL is tolerated by
GREED, but not by OCP.
So far, we have introduced four constraints in this analysis. Given the tendency in (3), the
following ranking should be expected. The language specific OCP constraint, given Panini’s
theorem, occupies a dominating position. The second most dominating constraint is a set of
ANCHORING constraints, which has a sub-ranking per se mapped from syntax. This constraint
punishes tonal shifts in right edges. Since the sub-ranking is directly mapped from syntax, the
tendency (the higher, the more faithful) is thus captured. The constraint in turn dominates GREED
because even in word domains, right edges are still preserved. The least dominant constraint is
the faithfulness IDENT-IO, which punishes any over-applications of TS. IDENT-IO being
tone). This indicates that the pressure to undergo TS must be from other sources. I leave this point open.
15
dominated by GREED, the only tolerated over-applications of TS are therefore the ones applied in
the minimal domains. Given a simple LL input, the rankings among the four constraints are
characterized as follows:
(14)
The interaction of the constraints in Mandarin7
a. OCP(L) >> IDENT-IO:
RL, LR, RR  LL
b. OCP(L)>> ANCHORING-IO (Si, Rt):
RL, RR  LL
c. ANCHORING-IO (Si, Rt) >> GREED(L, ΣMIN ):
<RL>  <RR>, <LR>
d. ANCHORING-IO (Si, Rt) >> IDENT-IO:
RL  LL
e. GREED(L, ΣMIN )>> IDENT-IO:
<RR>  <LR>, <RL>  <LL>
(15)
OCP(L)
ANCHORING-IO (Si, RT)
mapped from syntax
GREED(L, Σ
MIN
)
IDENT-IO
The Hasse diagram in (15) shows how this system works. In the next section, I apply the current
ranking in the analysis of Mandarin TS.
3.2
TS in Mandarin
3.2.1 The Directional Asymmetry in Compounds
As a first example, a simple LL sequence is put to test. The proposed ranking correctly predicts
that we have a RL sequence as the optimal candidate:
7
One question here is what ensures that the Low tones always change to the Rising tone. I do not explore the issue
in this paper, but only assume that there is a dominating constraint which avoids the Low tone to become other tones
than the Rising tone. The same assumption will also be made regarding TS in Taiwanese. However, this is an
interesting issue and I will leave it for future research.
16
(16)
<LL>1
a. N: zong-tong
‘president’
b. V: guan-li
‘to manage’
L L
UT
R L
ST
c. Visible String by TS:
S1 = <LL>1
<LL>1
a.
LR
b. → RL
c.
RR
d.
LL
OCP(L)
ANCHOR-IO (Si,RT)
S1
*!
GREED(ΣMIN)
S1
*
*
*!
*!
IDENT-IO
*
*
**
**
Since we have only one phonological string, S1, the ANCHORING constraint evaluates the right
edge of this string. Being a least dominant string itself, S1 is also evaluated by GREED(ΣMIN). The
result, as in (16), is that RL is the optimal output.
With respect to the directional asymmetry in compounds, the current analysis
immediately provides an explanation. The asymmetry between left and right compounds comes
from the competitions among the ANCHORING-RIGHT constraints (targeting different syntactic
objects), which compete with each other to preserve its right edges, and also from GREED(ΣMIN),
which drives exhaustive TS inside the minimal domains unless punished by the
ANCHORING-RIGHT. Examples (17)/(18) and (19)/(20) illustrate this asymmetry:
(17)
<<LL>1L>2
a. N: Zong-tong=fu ‘presidential office’
b. Visible Stings (minimal rhythmic unit: two tone bearing units (TBU))
S1 = <LL>1 zong-tong
S2 = <LLL>2 zong-tong=fu
c. Phonology-Syntax Mapping: S2 > S1
17
(18)
<<LL>1L>2
OCP(L)
ANCHOR-IO(Si ,RT )
S2
S1
a. → RRL
(19)
b.
LLL
c.
LRL
d.
RLL
e.
RRR
f.
LLR
g.
RLR
h.
LRR
GREED(ΣMIN) IDENT-IO
S1
*
*!*
*
*!
*!
*!
*
*!
*!
**
**
*!
*
*
*
**
*
*
*
*
***
*
**
**
<L<LL>1>2
a. N: lao=zong-tong ‘the senior president’
b. Visible strings by TS
S1 = <LL>1
zong-tong
S2 = <LLL>2
lao=zong-tong
c. Mapping: S2 > S1
(20)
<L<LL>1>2
OCP(L)
ANCHOR-IO(Si ,RT )
S2
a.
RRL
b.
LLL
S1
RLL
e.
RRR
f.
LLR
*!*
g.
RLR
h.
LRR
*!
*!
S1
*
**
*
**
c. → LRL
d.
GREED(ΣMIN) IDENT-IO
*!
*
*!
*!
*
*
*
*
*
*
**!
*
*
***
*
**
**
We have a rightward compound in (17)/(18) and a leftward compound in (19)/(20). The minimal
rhythmic unit effect on TS (Chen 2000) filters out the possibility that a single L will hold a
phonological string visible to the ANCHORING and GREED constraints (since TS is triggered by at
least two TBUs; the minimal rhythmic unit effect bears similarities to the minimal word effect in
stress-prominent languages; see Hayes 1995). In both cases the least dominant string is S1.
18
GREED targets S1, and in (17), the exhaustive TS in S1 does not result in any change of the right
edge of the higher domain (S2). At the same time, it avoids the OCP violations. Therefore, the
greedier candidate (a) in (18) wins over the modest candidate (18c). This is not the case in
(19)/(20), where we have a leftward compound. The exhaustive TS driven by GREED in S1
implies changing the right edge of the higher S2 (since S1 and S2 overlap in the right edge).
Therefore, candidate (e) and (g) will be filtered out. This time, the optimal candidate (20c) wins
over candidate (20a) since the latter involves over-application of TS outside the minimal
domains (and therefore is ruled out by IDENT-IO).
3.2.2 TS in Head-Complement Constructions
In the sentence levels, let us first examine head-complement constructions across different
syntactic categories. With respect to the verbal domain, (21a) and (23a) both involve a verb and
its NP complement; as for the nominal domain, (21b) and (23b) demonstrate cases with a
determiner or a classifier and its NP complement. As for PPs, (23c) involves a preposition and its
NP complement.
Since the same dominance relations are shared across word and sentence levels, our
formulations of phonology-syntax mappings predict that the same directional asymmetry will be
displayed in sentence levels. The contrast between (21)/(22) and (23)/(24) confirm the prediction.
In addition, the case in (25) shows when GREED targets two equally least dominant domains, the
TS patterns in the two domains will come out the same, as desired:8
The same pattern is seen in the 2+2 compound, such as lao-hu=qiu-zhang ‘tiger chief’, which has the underlying
tones <<LL><LL>> and the surface tones <<RL><RL>>.
8
19
(21)
[L[<LL>1]2]3
a. VP:
qing Lao-li
b. DP:
you wu-shou
c. Visible Strings by TS
S1=<LL>
S2= [LL]
S3= [LLL]
d. Mapping: S3 > S2 > S1
‘ask Laoli
‘there are five CL.’
(22)
[L[<LL>1]2]3
OCP(L) ANCHOR-IO (Si,RT) GREED (ΣMIN)
S3
S2
S1
S1
a. → LRL
b.
RRL
c.
RLL
d.
LRR
e.
LLR
f.
RLR
*
*
**
*!
*!
*
*!
*!
*
*
*
IDENT-IO
*
*
*
*
*
*
**!
*
**
*
**
(23) [[<LL>1]2L]3
a. VP:
gan-zou ma ‘expel the horse’
b. CLP:
wu-liang jiu ‘five CL. wine’
c. PP+AP:
bi-ma
lao ‘older than the horse’
d. Visible Strings
S1 = <LL>
S2 = [LL]
S3 = [LLL]
e. Mapping: S3 > S2 > S1
(24)
[[<LL>1]2L]3
OCP(L)
ANCHOR-IO (Si ,RT)
S3
a.
LRL
b. → RRL
c.
RLL
d.
LRR
e.
LLR
f.
RLR
S2
S1
S1
*
*
*
*
*!
*!
*!
GREED (ΣMIN)
*
*
*!
*!
20
*
*
**
*
IDENT-IO
*
**
*
**
*
**
(25) [[<LL>1]3[<LL>2]4]5
a. CLP: wu-shou hao-qu
‘five CL. good songs’
b. VP: guan-li zhi-chang
‘manage the paper factory’
c. Visible Strings
S1 = <LL>
‘guan-li’
S2 = <LL>
‘zhi-chang’
S3 = [LL]
‘guan-li’
S4 = [LL]
‘zhi-chang’
S5 = [LLLL]
‘guan-li zhi-chang’
d. Mapping:
S5 > S4,S3 > S2,S1
[[<LL>1]3[<LL>2]4]5
OCP(L)
GREED (ΣMIN)
ANCHOR-IO (Si ,RT)
S5
S4
S3
S2
S1
S1,S2
*,*
*,*
**,**
a. → RLRL
b.
LRLR
c.
LLLL
d.
RRRR
e.
RLLL
f.
RRRL
g.
RRLL
h.
RLLR
i.
LRRL
j.
LRRR
k.
LLRR
*!
*
*
*
*
*!
*
*
*
*
*!*
*!*
*!
*!
*!
*!
*
*
*!
*
*
*
*
*
*!
*
*
*
*
*
*
*
*,**
*
**
*,*
*,*
*
**
IDENT-IO
**
**
****
*
***
**
**
**
***
**
The optimal candidate (a) is the most balanced output between the pressure of preserving the
edges and applying exhaustive TS in the minimal domains.
3.2.3 Manner Adverbs
It is observed that lower (manner) adverbs universally have a close relationship to the verb (see
Costa 2004; Jackendoff 1972; Rivero 1992). In Mandarin, I assume that a manner adverb
syntactically forms a constituent with a verb (by syntactic incorporations or morphological
21
mergers).9 A simple coordination test gives evidence to the structure. A verb can form a
constituent with a manner adverb, as in (26a) and (26b), but not with a higher adverb, as in (26c)
and (26d), where we have a temporal adverb and a modal adverb, respectively:
(26)
a. Zhangsan [[luan
nian] you
[luan
hua]
na ben shu].
Zhangsan casually read and
casually scribble that CL book
‘Zhangsan read and scribbled on that book in a casual manner.’
b. Zhangsan [[suibian zhengli] you [ suibian mai]
tade dongxi].
Zhangsan
sloppily arrange and sloppily sell
his stuff
‘Zhangsan arranged and sold his stuffs in a sloppy manner.’
c. *Zhangsan zuotian
nian-le you
zuotian
hua-le
na ben shu
Zhangsan yesterday read and
yesterday scribbled
that CL book
‘intended: Zhangsan read and scribble on that book yesterday.’
d. *Zhangsan dagai
mai-le you
dagai
mai-le tade dongxi.
Zhangsan probably bought and
probably sold
his stuff
‘intended: Zhangsan probably bought and sold his stuffs.’
Given the attested structure of a manner adverb, the proposed ranking successfully selects correct
output tonal pattern. See (27)/(28):
9
Alternatively, Hagit Borer (pc) points out that the structure may as well be a result from VP raising. If we assume
that object is licensed outside the VP (Huang 1997; Lin 2001; Borer 2002), and the manner adverb is licensed in
SpecVP, then a Kayne-style VP movement results in the observed structure. Whatever the syntactic analysis might
be, the structure will be the same as in (27).
22
(27)
[[[<LL>1]2L]3[<LL>4]5]6
a. VP: hao-hao qing
zong-tong
well
ask
president
‘ask the president in a good manner’
b. Visible Strings:
S1 =
<LL>
‘hao-hao’
S2 =
[LL]
‘hao-hao’
S3 =
[LLL]
‘hao-hao qing’
S4 =
<LL>
‘zong-tong’
S5 =
[LL]
‘zong-tong’
S6 =
[LLLLL] ‘hao-hao qing zong-tong’
c. Mapping:
S6 > S3, S5 > S2, S4 > S1
(28)
[[[<LL>1]2L]3[<LL>4]5]6
OCP(L)
S6
a.
RLRLL
b.
RRLLL
*!
*!,*
S3
S5
RLLRL
e.
RLRLL
f.
RLRRL
g.
RRRRL
h.
LRLRL
i.
LRRRL
j.
RRRRR
S2
S4
S1
*
c. → RRLRL
d.
Greed(ΣMIN)
Anchor-IO (Si, Rt)
*!
*!
*
*!
*!
*!
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
IDENT-IO
S1,S4
*,**
**
*
*,*
*
*,*
*
*,*!
*,*!
**
**
***
**
**
***
****
**
***
*****
3.3 The Residual Problems: Subjects and Higher Adverbs
The directional asymmetries in compounds and head-complement structures in Mandarin TS are
explained by the proposed ranking. The proposed ranking, however, runs into problems as soon
as subjects and higher adverbs are brought into attentions. Take the following sentence for
example, where a subject is present:
23
(29) zong-tong
president
L L
R R/L
?R
L
qing
invite
L
L
R
Lao-li
Laoli
L L
R L
R L
UT
ST1/2
ST3 (marked)
TP, Σ1
NPi, Σ4
T’
VP, Σ2
T
N, Σ5
ti
V’
NP, Σ3
V
N, Σ6
zong-tong
[[<LL>5]4[L[<LL>6]3]2]1
qing
Lao-li
OCP(L)
S1
a. RLRLL
*!
b. RRLLL
*!,*
Greed(ΣMIN) IDENT-IO
Anchor-IO (Si, Rt)
c.  RRLRL
S4
S2
S3
S5
S6
S5,S6
*,**
**
*
*
**
**
*!
*
*
***
d.  RLLRL
*!
*,*
**
e. RLRLL
*!
*,**
**
*,*
***
*
****
f. → RLRRL
g. RRRRL
h. RRRRR
i. LRRRL
*!
*!
*
*
*
*!
*
*
*
*
*****
*,*
***
The proposed ranking incorrectly selects the marked candidate (f) as the optimal output, but the
desired output is actually either (c) or (d). The same problem occurs when we deal with TS of
higher adverbs. For convenience, I ignore the tableaux here. As mentioned before, the problem is
that we have an optional TS boundary between subject and VP, and between a higher adverb and
24
VP (if we have a covert subject). The question is how these optional TS boundaries are derived,
and why they can only occur in such positions. The solution to this problem, I suggest, actually
lie in a broader linguistic context. In the next section, I show that TS in Taiwanese may provide a
clue to this problem.
4.
TS in Taiwanese: Prosodic Phrase by Multiple Spell-Out
4.1
Basic TS patterns
Taiwanese has five tones in total, and each tone undergoes TS in the combinations with any tone
(including itself). The tones in Taiwanese shift in a cyclic manner (generally referred to as the
tone cycle), illustrated in the following diagram (modified from Chen 2000: 432):
(30) Taiwanese tone cycle
a. in open syllables:
24(Rising)
22(Low)
44(High)
21(Low-falling; Lf)
53(High-falling; Hf)
b. in closed syllables:
44(H)
22(L)
Like in Mandarin, Taiwanese also preserves the underlying tones in the right edges. However, TS
in Taiwanese applies iterative all the way to the leftmost edge unless a TS boundary is met:
25
(31)
word-level
a. ten
electricity
H
L
sii
view
H
H
b. tsui ten
water electricity
Hf H
H
H
(32)
‘television’
UT
ST
‘utilities’
UT
ST
sentence-level
a. tsong-thong koŋ
president
speak
Hf Hf
Hf
( H Hf ) # ( H
‘The president told a lie.’
pee-tshat
white-thief
H L
L L)
UT
ST
b. tsong-thong ban-ban
gong
president
slowly
speak
Hf Hf
H
H
Hf
(H
Hf) # ( L
L
Hf )
‘The president spoke slowly.’
[lower adverb]
c. mi-a-tsai
kih
tai-pak
tomorrow
go
taipei
R Hf Lf
L
R H
( L H Lf ) # ( H
L H)
‘Go to Taipei tomorrow’
[higher adverb]
UT
ST
UT
ST
d. Tao-dua tsong-thong boh
lai
just
president
not
come
H Hf Hf Hf
R
Hf
UT
( L H H Hf )# ( L
Hf )
ST
‘The president did not come not very long ago.’
26
A descriptive generalization concerning Taiwanese TS boundaries is made by Chen (2000):
(33)
P-phrasing in Xiamen
{Right, Xmax}, where Xmax is not an adjunct
The generalization makes correct predictions with respect to the subject-verb and verb-object
cases (since a subject DP is an Xmax, but an object DP is embedded in VP, no Xmax is found
between V and object), but the asymmetry of adverbs is not accounted for. For example,
although the generalization correctly predicts that in (32b) a TS boundary does not occur in
between the manner adverb and verb, however, the temporal adjunct unexpectedly introduces a
TS boundary in (32c).
Upon close scrutiny, the data from Taiwanese TS are reminiscent of the problematic
asymmetries in Mandarin TS regarding subjects and higher adverbs. The data in (32) further
suggest that a sentence seems to be split into several major groups, in which TS applies
independently (insensitive to elements in other major groups). Abstracting the structures, we find
that the obligatory TS boundaries in Taiwanese coincide exactly with the optional TS boundaries
in Mandarin:
(34)
The Major Grouping Effect
Temproal Adv.
a. Taiwaense:
b. Mandarin:
Subject
*#
*#
Major grouping:(
[VP M.Adv.
Verb
Object]
#
*#
*#
(#)
*#
*#
)(
)
In general, there appears to be a forced boundary immediately before the main VP (as we take
manner adverbs a part of VP). In the following, I argue that this boundary is introduced by
multiple spell-outs.
27
4.2
Multiple Spell-outs and Taiwanese TS
One possible account for the ‘major grouping effect’ is to say that these major groups come from
multiple spell-outs. Since Chomsky (1995), the concept of spell-out is introduced to syntactic
theory. The idea of spell-out is that during a syntactic derivation, the phonological parts of the
structure are taken to PF at a certain point, called spell-out. Along the same line, Chomsky (2000,
2001) further develops a radical version of the idea in the Strong Minimalist Thesis, which
incorporates multiple spell-outs to account for cyclicity, a core property of syntax. The rationale
behind multiple spell-outs is that when a syntactic derivation reaches a certain point, it is casted
to PF. Therefore, the later syntactic operations can no longer retrieve information from spelt-out
domains. On the one hand, multiple spell-outs explain why syntax resists any countercyclic
operations; on the other, multiple spell-outs also provide a blue print for the syntax-phonology
interface, since each spell-out ships some information to phonological operations. The proposal
of multiple spell-outs has been used to analyze several phonology-syntax mapping asymmetries;
especially issues concerning syntactic domains for phonological operations (see Simpson & Wu
2002 for Taiwanese IP-raising and TS of the verbal complementizer gong; Ishihara 2004, 2007
for intonation patterns in Japanese; Kratzer & Selkirk 2007 for prosodic stress patterns in
German).
Following Chomsky (2000, 2001), I assume multiple spell-outs work in the following
way. When a derivation builds a phase, the phase spell-outs to PF except its edges (head and
Spec), which are accessible to the later derivations. (35) gives us an idea of how multiple
spell-outs restrict syntactic shipments to PF. Each spell-out automatically generates the so-called
‘major phrases’ (Kratzer & Selkirk 2007) for phonological operations. Not surprisingly, the
major phrases coincide with the TS boundaries in Taiwanese:10
10
Given the fact that agent-oriented adverbs are also phrased together with the main VP, I assume that the spell-out
28
(35) Phases and Spell-outs
CP/FocusP
spell-out of CP phrase
SpecCP
C
TP
AdvP TP
DP
T
vP
spell-out of vP phrase
Subject
SpecvP
v
VP
AdvP
V
DP
Object
(
)#(
)#(
) # : major phrases
If TS domains are ultimately restricted by cyclic spell-outs, the puzzling asymmetries in
Mandarin and Taiwanese TS can be straightforwardly accounted for. We can independently
assume that a major phrase always introduces an obligatory TS boundary in Taiwanese, but TS is
optional between the major phrases in Mandarin; therefore, an LL sequence is only tolerated
between the edges of major phrases.11
domain in Taiwanese might be vP, instead of VP (if such adverbs are adjoined to vP). The same proposal has been
claimed in Korean by Ko (2007) and in Japanese by Takita (2007). Another possibility is that agent-oriented adverb
is adjoined to the main VP.
11
It seems that the major TS boundaries can be ignored only in fast speech. If the observation is right, then these
major phase boundaries are also obligatory in Mandarin, but can be cancelled by other prosodic factors. I leave this
point pending.
29
With the help of the cyclic spell-outs, we can also extend our analysis to cover Taiwanese
TS with the same set of phonology-syntax mapping constraints. All we need is to change the
parameters in the mapping constraints. First, in contrast to that in Mandarin, GREED in Taiwanese
TS targets the maximal strings (given by a major phrases). Second, the ANCHORING constraint
also targets on the maximal syntactic domains. The proposed ranking in Taiwanese is as follows:
(36)
Ranking for Taiwanese TS
Anchoring (ΣMAX, R) >> Green (ΣMAX) >> IDENT-IO
The ranking in (36) correctly predicts exhaustive TS in a major phrase given by a spell-out. Take
the following structure in (37) for example. For convenience, I use X to represent the underlying
tones and Y the shifted tones (X and Y are tonal variables). Since the most dominant string is S1,
the greediest candidate that preserves the right edge in S1 is selected, which is candidate (g):
(37)
[[<XX>4X]3[<XX>5]2]1
Anchor-IO(SMAX,R ) Greed (ΣMAX)
S1
a.
YXYXX
b.
YYXXX
c.
YYXYX
d.
YXXXX
e.
YXYXX
f.
YXYYX
S1
**!*
**!*
**!
**!**
**!*
**!
*
**!*
g. → YYYYX
h.
XYXYX
i.
YYYYY
IDENT-IO-T
*!
**
**
***
*
**
***
****
**
*****
Actually, the complexity of the syntactic structure in (37) does not weigh influence on the
surface tonal pattern. The constraints only scan the maximal string(s) held by the most dominant
syntactic category. We will see the same point when tone grouping in Tohono O’odham is
30
analyzed in 5.2.
5.
Tone Sandhi in Other Languages
In this section, I examine the universal aspect of the proposed direct phonology-syntax mapping
hypothesis. First, I extend my analysis to cover Shanghai Chinese tone spreading. Although the
surface tonal patterns look drastically different from those in Taiwanese and Mandarin, I show
that the same set of phonology-syntax mapping constraints allow us to capture tone spreading in
Shanghai Chinese, which is only minimally different from Mandarin TS by language specific
constraints. Then I try to give a reanalysis on tone grouping in Tohono O’odham, which, I argue,
functions like Taiwanese TS in targeting the maximal domains.
5.1
Tone Spreading in Shanghai Chinese
In Shanghai Chinese, TS works quite differently from Taiwanese and Mandarin. Instead of
shifting the underlying tones in a fixed manner, Shanghai Chinese involves tonal spreading of
contour tones within certain domains. Usually, tone spreading applies obligatorily inside word
domains. The leftmost contour tones (such as XY) are spread across a word domain: an adjacent
word will receive the right part of the contour tone (i.e. Y), and a nonadjacent word will be
assigned a default low tone (Ldef). Examples of tone spreading can be found in (38) (from Selkirk
& Shen 1990: 314):
(38)
a. thi - tshi
H1L2
MH
(H1
L2 )
‘weather’
UT
ST
b. hu M1H2
(M1
‘anger’
UT
ST
tshi
MH
H2)
31
c. wã - tso. zo
L1H2 MH LH
(L1 H2 Ldef)
‘professor Wang’
UT
ST
d. baq.çoq-kõ. tsï
LH.MH HL.MH
(L H) (H L)
‘white snow princess (Snow White)’ (Chen 2000: 311)
UT
ST
(from Duanmu 1993:21)
Another type of tone spreading applies across separate word domains. The spreading
specifically targets on function items. I call this type of spreading ‘function word spreading.’ A
minimal contrastive pair is shown in the following. The feature [+F] indicates that an element is
a function item (also from Selkirk & Shen 1990: 317):
(39)
a. (V
tã
M1 H 2
( M1
pron.
‘nõ[+F]
LH
H2
Asp.)
leq[+F]
LH
L(def) )
‘has hit you.’
UT
ST
b. (V
tã
M1 H 2
M1H2
N
‘mo
L3H4
(L3
Asp.)
leq[+F]
LH
H4 )
‘has hit the horse’
UT
ST
In (39a), the verb is followed by two function words, each of which triggers tone spreading from
the nearest lexical item to its left. Applying the tone spreading allows the adjacent pronoun nõ
‘you’ to acquire the high tone (H2) from the verb tã ‘hit’, while the aspectual marker, which is not
adjacent to the verb, is specified with a default low tone. On the other hand, the only function
32
word in (b) is the Perfect aspectual marker leq, and therefore, tone spreading forces the spreading
of the high tone (H4) from the adjacent ‘mo ‘horse’ to the aspectual marker.
Let us see how we can capture tone spreading in Shanghai Chinese within the proposed
phonology-syntax mapping condition. First, we need to formulate a set of language specific
constraints for Shanghai Chinese to deal with the function word spreading. The constraint
F-SPREAD will be used:
(40)
F-SPREAD
A function word will either (i) take a tone from the nearest lexical item, or (ii) be
specified with a default low tone if not adjacent to a lexical item.
(when (i) or (ii) is not met, assign a * violation)
The word-internal spreading, as in Mandarin, is dealt with by GREED (ΣMIN), which targets the
least dominant domain (presumably the X0-level). When a domain contains more than one TBU
(i.e. a syllable/prosodic word, which always bears at least one and at most two tones in Shanghai
Chinese; I assume the minimal rhythmic unit effect requires tone spreading applies to at least two
TBUs), GREED punishes any faithful non-spreading strings. But if the least dominant domain
contains only one TBU (usually bearing two tones), then it is immune to GREED (since spreading
will be inapplicable), due to the minimal rhythmic unit effect.
As for the ANCHORING constraint, unlike in Mandarin, the ANCHOR-IO(Si, LEFT )
constraint in Shanghai Chinese targets left edges, but like the tendency in Mandarin, the higher a
domain is, the more likely it preserves its left edge; therefore, the ranking of the split
ANCHORING constraints is also directly mapped from syntax. The hierarchical mapping also
avoids reversed tone spreading. As for the final ranking in Shanghai, it is still similar to the
ranking in Mandarin: the language specific F-SPREAD dominates the ANCHORING-LEFT
33
constraints, which in turn dominates GREED (ΣMIN), and then IDENT-IO:
(41)
Ranking in Shanghai Chinese
F-SPREAD >> ANCHORING (S, LEFT) >> GREED (ΣMIN) >> IDENT-IO
Applying the ranking to examples in Shanghai, first, the contrast between (39a) and (39b) is now
captured by the high priority of the function word spreading:
(42)
a. (V
taN
MaHb
Ma
pron.
‘noN[+F]
LH
Hb
b.
AspP, S5
Asp.)
leq[+F]
LH
L(def)
‘has hit you.’
UT
ST
S4
V,S1
pron,S2
σ
X
σ
Y .X
Asp,S3
σ
Y X Y
(again, X and Y represent tone variables)
(43)
[[<XaYb>1. <Xc Yd>2]4. <XY>3]5
F-SPR
ANCHORING (S, LEFT)
S5
a. → (Xa. Yb. L)
b. (Xa. Yb). XY
c. XY. XY. XY
d. XaYb. (Xc. Yd)
e. L. L. L
f. XaYb. L. L
g. Yb. Xa. L
S4
*!
*!*
*!
*!
*
*!
*
*!
34
S1,S2,S3
GREED (ΣMIN)
IDENT-IO
S1,S2,S3
*,*
*
****
**
*,*
*,*,*
*,*
*,*,*
**
******
****
****
Candidates (b), (c), (f) respectively represent cases where functional spreading does not apply to
the aspectual marker, does not apply at all, or is replaced by default tone insertion. These
candidates are ruled out by the high-ranked F-SPREAD. Candidate (d) is when functional
spreading applies from a function item to another function item. Again, F-SPREAD rules it out
since the spreading does not apply to first function item. Candidate (e) is when every tone is
deleted and replaced by a default low tone. In such a case, no violations are assigned to
F-SPREAD because no tone can feed spreading. But this case will cause violations of the
Anchoring constraints.12 Candidate (g) shows reversed tone spreading is disfavored against the
ANCHORING-LEFT constraints. Note that GREED is not violated by either of the cases since each of
the minimal string contains only one TBU (hence immune to GREED). The opposite case is
shown in the following:
(44) a. (V
tã
MaHb
MaHb
b.
N
‘mo
LcHd
Lc
Asp.)
leq[+F]
LH
Hd
‘has hit the horse’
UT
ST
AspP, S5
S4
X
V,S1
N,S2
Asp,S3
σ
σ
σ
Y .X
YX
Y
12
I assume that F-SPREAD can also rule out a candidate like [Xa. Xc. X], in which every left edge is preserved
without any applications of functional word spreading.
35
(45)
[[<XaYb>1. <Xc Yd>2]4. <XY>3]5
F-SPR
ANCHORING (S, LEFT)
S5
a.
(Xa. Yb. L)
b.
(Xa. Yb). XY
c.
XaYb. XcYd. XY
S4
e.
L. L. L
f.
XaYb. L. L
S1,S2,S3
*!
*
*!
IDENT-IO
S1,S2,S3
*,*!
*
****
**
*
*,*,*
*,*
**
******
****
*!
*!
d. → XaYb. (Xc. Yd)
GREED (ΣMIN)
In this case, candidate (d) is selected as the optimal output. The over-application of functional
spreading in candidate (a) is ruled out by ANCHORING constraints, as desired.
So far, GREED(ΣMIN) has little to do with the filtering process. This is because both
examples involve monosyllabic words. In the cases where we have disyllabic (or more) words,
GREED(ΣMIN) will be involved. The following example is a case where we need both F-SPREAD
and GREED since we have both a function item and a disyllabic word:
(46)
a. (V
Prep
tsou taw[+F]
MaHb MH
Ma Hb
NP)
noe-tsĩ
LcHd HL
Lc
Hd
‘walk to Nanjing’
UT
ST
b. [VP <tsou>3 [PP<taw>4 [NP<noe-tsĩ>6]5]2]1
c. Mapping: S1 > S2, S3 > S4, S5 > S6
d. ΣMIN : S3, S4, S6
36
[ <AaBb>3 [<AcBd>4 [<AeBf. AB>6]5]2]1
F-SPR
ANCHORING (S, LEFT)
S1
a. → (Aa.Bb).(Ae.Bf)
b.
AB. AB. AB. AB
c.
(Aa. Bb). AeBf. AB
d.
AaBb. AcBd.( Ae.Bf)
S2, S3
S4,S5
*
*
S6
*!
GREED (ΣMIN)
IDENT-IO
S3, S4, S6
****
*
*
*
*!
*!
**
**
In the above tableaux, the optimal candidate (a) applies both types of tone spreading rules,
function word spreading and word-internal spreading. Candidate (b) is a case where no spreading
has taken place, therefore ruled out by F-SPREAD. Candidate (c) only applies function word
spreading, but not word-internal one, and candidate (d) the opposite. In both (b) and (c),
GREED(ΣMIN) kicks in since the minimal domain (S6) satisfies the minimal rhythmic unit
requirement.
5.2
Tone Grouping in Tohono O’odham
On surface, tone grouping in Tohono O’odham (TO) displays quite different properties from
Mandarin/Taiwanese TS or Shanghai tone spreading. However, in the following I argue that,
except for language-specific constraints, the phonology-syntax mapping part in TO is on the
same par as that in other tonal languages (all of the following examples in TO are drawn from
Hale & Selkirk (1987) and Truckenbrodt (1999)).
Tone grouping in TO can be simply characterized as follows: (i) each tone group is
n
n
n
evenly distributed with a skeletal tonal pattern [L 0 H 1 L 1] (i.e. a series of high tones can be
preceded by zero or more low tones, and followed by at least one low tones), and (ii) the series
of high tones begins with the first stressed vowel and ends in the final stressed vowel within a
tone group. For example, consider the following hypothetical sequence:
37
(47)
a. bá.ba.bá.bá.ba.ba
( H. H. H. H. L. L ) #
b. ba.bá.ba.bá.ba. bá.ba.ba
( L. H. H. H. L ) # .(H. L. L )
(47a) shows that in a single tone group, the consecutive high tones extend from the leftmost
stressed vowel to the rightmost one. The rest unstressed syllables are inserted with a default low
tone. In (47b) there are two tone groups. In the second group, there is only one stressed vowel (it
is the rightmost as well as the leftmost stressed vowel per se). Therefore, only one high tone is
found in this tone group.
Since the tonal skeleton is rather fixed, the problem in TO tone grouping lies on how to
find out the correct tone group boundaries. Here I review three examples from Truckenbrodt
(1999):
(48)
a.
IP
DPi
I’
I
VP
ti
V’
DP
wákial ’at
g wísilo
cowboy Aux
the calf
( H L L) #(L
HHH
or ( H L L L ) # (
HHH
V
cépos
branded
‘The cowboy branded the calf.’
HL ) #
HL ) #13
13
Truckenbrodt does not give the latter phrasing possibilities, which does not seem to violate any skeletal
requirements. In fact, given multiple spell-outs, I argue this is the correct phrasing.
38
b.
IP
I
VP
DP
V’
DP
na-t
(L
g wákial
HHH
c.
V
g wísilo
HHH
cépos ‘Did the cowboy brand the calf?’
HL)#
IP
I
VP
VP
DP
V’
ti
na-t g wákial
(L
HHH
DPi
V
cépos
g wísilo
H L ) # (H L L) #
’Did the cowboy brand the calf?’
These examples immediately remind us of the subject/object asymmetry in Taiwanese. That is, in
(48a), a tone boundary is introduced in between subject and VP (when subject raises to SpecIP).
On the other hand, in example (48b), where subject remains in VP, the whole sentence is inside a
single tone group. Similar to Taiwanese, in (48c) object shift (right adjunction to VP) also
introduces an independent tone group. These examples reveal that tone grouping targets the most
dominant strings as TS in Taiwanese. Similarly, due to multiple spell-outs, a major phrasing is
forced in between subject and VP. On the other hand, in the object right-dislocation structure, by
definition of dominance (i.e. X dominates Y only if all segments of X dominate Y), the higher
VP segment does not dominate the right-dislocated object DP in (48c). Therefore, object DP and
39
VP do not dominate each other -- they are equally dominant when VP is spelt-out to PF.
Therefore, each DP and VP holds a maximal string. Since tone grouping targets only the
maximal strings, the proposed theory successfully predicts that two independent tone groups will
be created through VP-adjunction.
Without exercising the tonal skeleton, we can actually reanalyze the tone grouping in
TO by restricting distributions of the high tones, and like Shanghai Chinese, surface low tones
are inserted by default. Therefore, the constraints needed to capture the tone grouping in TO are
formulated as follows:
(49)
WRAP-ΣMAX
A most dominant string is contained in a single tone group.
(50)
*(HLH)#
A low tone cannot appear in between two high tones within a tone group.
(51)
ALIGN (H, Left/Right, V[+stress] in ΣMAX)
A high tone coincides with the leftmost stressed vowel and with the rightmost stressed
vowel within a maximal string.
(52)
*H
High tones are marked.14
(49), (50), and (51) together shape the tonal skeleton in TO. (49) is inspired by Truckenbrodt’s
(!999) analysis, but instead of having WRAP targeting XP, I assume that WRAP is another
phonology-syntax mapping constraint, and therefore, it is able to target ΣMAX. (50) and (51)
account for the fact that a series of high tones will be aligned with the stressed vowels, and (50)
creates the pressure against a low tone from intervening. The markedness constraint *H, on the
other hand, prevents this language uses too many high tones (unless they are forced by the
14
Another way of formulating this constraint is that we assume that all syllable bear low tones in the input, and
IDENT-IO (L) will have same effect. *H is adopted here just for simplicity.
40
ALIGNMENT and other higher-ranked constraints). The following minimal pair illustrates how
these constraints work. Note that in (48b’) (cf. (48b)), the Auxiliary head itself is grouped in a
tone group due to multiple spell-outs, and it is inserted with a default low tone (because it bears
no stress).15
(48)
b’. na-t g-wákial
g-wísilo cépos
Aux. the.cowboy
the.calk branded
( L )# (H H H
HHH
HL)#
‘Did the cowboy brand the calf?’
(53)
[<g-wákial> [<g-wísilo> <cépos>]2]1
*(HLH)#
WRAP-ΣMAX
ALIGN (H, L/R,
V[+stress] in Σ
MAX
S1
HHH
HHH
*******
HL )#
b. [<g-wákial> [<g-wísilo> <cépos>]2]1
(
HHH
HHL
LL
*!
HLL )#
HHH
*!
*!
******
HL )#
e. [<g-wákial> [<g-wísilo> <cépos>]2]1
(
HHH
HHH
*****
(HL )#
d. [<g-wákial> [<g-wísilo> <cépos>]2]1
( HHL
*****
)#
c. [<g-wákial> [<g-wísilo> <cépos>]2]1
( HHH
)
S1
a. → [<g-wákial> [<g-wísilo> <cépos>]2]1
(
*H
*******
HH )#
*!
Candidate (a) is the optimal output although it incurs many violations on *H, which are driven
by the higher ranked alignment constraint and the *HLH constraint. Candidate (b) shows a
situation when high tones are not placed in the correct positions (the rightmost stressed vowel is
not aligned with a high tone). If we try to save candidate (b) by reassign the tone group, as in
candidate (c), it will still be ruled out because the grouping will necessarily incur violations of
WRAP-ΣMAX. Candidate (e) shows a case of overapplying the high tone assignment. As expected,
15
Through multiple spell-outs, we also derive the lexical category condition. The asymmetry between lexical and
functional categories results from the independent spell-out of lexical and functional phases.
41
the *H constraint rules it out.
(48)
c’. na-t g-wákial
( L )# ( H H H
cépos
g-wísilo
H L ) # (H L L) #
‘Did the cowboy brand the calf?’
(54)
[<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
*(HLH)#
WRAP-ΣMAX
ALIGN (H, L/R,
V[+stress] in Σ
MAX
S1, S2
*H
)
S1, S2
a. → [[<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
HHH
H L )#
( H L L )#
b. [<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
HHL
HL
)#
*****
*!
****
(H L L )#
c. [<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
LHH
HH
LHH
HH
*****
*!
*****
H L L )#
d. [<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
*!
)#
( HLL )#
e. [<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
HHH
H H )#
******!
(H L L )#
f. [<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
H HH
HH
******!
H L L )#
g. [<g-wákial>4 [<cépos>]3]2 [<g-wísilo>]1
(
HLL )#
(H L) #
*! (S2)
***
(H L L)#
The object right-dislocation creates two maximal strings (S1 and S2) at the point of the lower
phase spell-out. By definition, the two strings are equally dominant. Therefore, Wrap-Σmax
predicts that either (i) both S1 and S2 are contained in the same tone group, or (ii) each S1 and
S2 is wrapped in the same tone group. As a result, (g) is ruled out: S2 is not contained in a single
tone group, but two. Given the two possible ways of grouping, candidates (c) and (d) both fail to
satisfy the alignment constraint since the stressed vowel in the right edge is not aligned with a
high tone. Moreover, even if the alignment is satisfied, as in candidates (e) and (f), the
markedness *H will rule them out. Eventually, the grouping (ii) is preferred, as in candidate (a)
42
since less high tones will be forced by the structure.
6. Conclusion
I propose that structural dominance is the single source for phonology-syntax mappings.
Rankings of the mapping constraints in phonology directly reflect the hierarchy in syntax. On the
other, a mapping constraint can also target either the most or the least dominant domains in
syntax. The two types of mapping relations are seen in TS in Mandarin and Taiwanese. In
Mandarin, the ranking of the ANCHORING constraints are imported from the structural hierarchy
and Greed targets the minimal domains, where in Taiwanese, both constraints apply only to the
maximal domains. We see Shanghai Chinese employ basically the same phonology-syntax
mapping strategies as Mandarin. Tohono O’odham, however, behaves more like Taiwanese with
respect to the mapping applications. In these languages, only the maximal strings mapped from
the most dominant syntactic domains are visible by the mapping constraints. If the proposed
analyses are on the right track, a universal mapping condition between phonology and syntax can
be successfully maintained.
43
REFERENCES
Alderete, John. 2001. Dominance effects as transderivational anti-faithfulness. Phonology
18:201-253.
Borer, Hagit. 2003. Exo-skeletal vs. endo-skeletal explanations: Syntactic projections and the
lexicon, in J. Moore and M. Polinsky, (eds.) The nature of explanation in linguistic theory,
31-67. Stanford, CA: CSLI Publications.
Chen, Matthew. 2000. Tone sandhi. Cambridge: Cambridge University Press.
Chen, Matthew. 1990. Why must phonology know about syntax? in Sharon Inkelas and Draga
Zec (eds.) The Phonology-Syntax Connection, 313-337. Chicago: University of Chicago
Press.
Cheng, Lisa. 1987. On the prosodic hierarchy and tone sandhi in Mandarin, in P. Avery (ed.)
Toronto Working Papers in Linguistics 7, 24-52. Linguistic Graduate Course Union,
University of Toronto.
Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 1995. Minimalist Program. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: The framework. In R. Martin, David Michaels, and
Juan Uriagereka (eds.) Step by Step: Essays in Minimalist syntax honoring Howard
Lasnik, 89-115. Cambridge Mass.: MIT Press.
Chomsky, Noam. 2001. Derivation by phase. in M. Kenstowicz (ed.) Ken Hale: a life in
language, 1-52. Cambridge, Mass.: MIT Press.
Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic perspective. Oxford:
Oxford University Press.
Costa, João. 2002. A multifactorial approach to adverb placement: Assumptions, facts, and
problems. Lingua 114: 711-753.
Duanmu, San. 1993. Rime length, stress, and association domains. Journal of East Asian
Linguistics 2: 1-44.
Hale, Kenneth and Elisabeth Selkirk. 1987. Government and tonal phrasing in Papago.
Phonology Yearbook 4: 151-183.
Hayes, Bruce. 1984/1989. The prosodic hierarchy in meter. In Paul Kiparsky and G. Youmans
(eds.) Rhythm and Meter, 201-260. Orlando: Academic Press.
Hayes, Bruce. 1995. Metrical stress theory. Chicago: University of Chicago Press.
Huang, C.-T. James. 1997. On lexical structure and syntactic projection. Chinese Languages and
Linguistics 3: 45-89.
Ishihara, Sinichiro. 2004. Prosody by phase: Evidence from focus intonation - wh-scope
correspondence in Japanese. Interdisciplinary Studies on Information Structure 1:
77–119.
Ishihara, Sinichiro. 2007. Major phrase, focus intonation, and multiple spell-out. The Linguistic
Review 24: 137-167.
44
Jackendoff, Ray. 1972. Semantic interpretations in generative grammar. Cambridge, Mass.: MIT
Press.
Ko, Heejeong. 2007. Asymmetries in scrambling and cyclic linearization. Linguistic Inquiry 38:
49-83.
Kratzer, Angelika and Elisabeth Selkirk. 2007. Phase theory and prosodic spellout: The case of
verbs. The Linguistic Review 24: 93-135.
Lin, Jonah. 2001. Light verb syntax and the theory of phrase structure. Ph.D. dissertation,
University of California, Irvine.
McCawley, James. 1968. Concerning the base component of a transformational grammar.
Foundations of Language 4: 243-269.
Prince, Alan and Paul Smolensky. 1993. Optimality theory. Technical Report no.2, Rutgers
Centers for Cognitive Science, Rutgers University.
Rivero, Maria Luisa. 1992. Adverb incorporation and the syntax of adverbs in Modern Greek.
Linguistics and Philosophy 15: 289–331.
Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. Phonology Yearbook 3:
371-405.
Selkirk, Elisabeth. 1995. Sentence prosody: intonation, stress, and phrasing. In John Goldsmith
(ed.) The Handbook of Phonological Theory, 550-569. London: Blackwell.
Selkirk, Elisabeth and Tong Shen. 1990. Prosodic domains in Shanghai Chinese. in Sharon
Inkelas and Draga Zec (eds.) The Phonology-Syntax Connection, 313-337. Chicago:
University of Chicago Press.
Shih, Chi-lin. 1986. The prosodic domain of tone sandhi in Chinese. Ph.D. dissertation,
University of California, San Diego.
Simpson, Andrew and Zoe Wu. 2002. IP-raising, tone sandhi and the creation of S-final particles:
Evidence for cyclic spell-Out. Journal of East Asian Linguistics 11: 67-99.
Takita, Kensuke. 2007. The proper binding condition effect as a consequence of cyclic
linearization. Paper presented in NELS 38, Ottawa University, Oct. 27 2007.
Truckenbrodt, Hubert. 1999. On the relation between syntactic phrases and phonological phrases.
Linguistic Inquiry 30: 219-255.
45